The standard library hash map implementation in C++ (std::unordered_map) is notoriously slow and memory-hungry (this cannot be fixed due to stricter iterator requirements on std::unordered_map).
As we use maps in a few places, and are likely to use them more in the future (e.g. for dynamically mapping IDs from the TSV), so it makes sense to switch to a better implementation.
I picked this one in particular because it seems to be one of the most resource-effective and well-maintained implementations at the moment. Additionally, it support heterogenous lookup, which is not yet available in std::unordered_map on some of the platforms that we support.
The last commit in this PR also switches the few std::map instances to the new container. Those maps didn't need to be ordered.
Switches to the excellent https://github.com/martinus/unordered_dense
Why?
The standard library hash map implementation in C++ (
std::unordered_map) is notoriously slow and memory-hungry (this cannot be fixed due to stricter iterator requirements onstd::unordered_map).As we use maps in a few places, and are likely to use them more in the future (e.g. for dynamically mapping IDs from the TSV), so it makes sense to switch to a better implementation.
I picked this one in particular because it seems to be one of the most resource-effective and well-maintained implementations at the moment. Additionally, it support heterogenous lookup, which is not yet available in
std::unordered_mapon some of the platforms that we support.The last commit in this PR also switches the few
std::mapinstances to the new container. Those maps didn't need to be ordered.